VOLUMETRIC computed tomography (CT) based on flatpanel

IEEE TRANSACTIONS ON NUCLEAR SCIENCE, VOL. 56, NO. 3, JUNE 2009 1179 Radiation Effects on the Performance of CMOS Photodiode Array Detectors and the Role of Gain-Offset Corrections Ho Kyung Kim, Min Kook Cho, Thorsten Achterkirchen, and Wonho Lee Abstract We report the observation of performance degradation in a detector consisting of a phosphor screen and a CMOS (complementary metal-oxide-semiconductor) photodiode array under the continuous irradiation of 45-kVp x-rays. The performance was assessed in terms of dark pixel signal, dynamic range, modulation-transfer function (MTF), noise-power spectrum (NPS), and detective quantum efficiency (DQE). From the measurement results, it has been observed that the increase of dark pixel signal and the related noise gradually reduces the dynamic range as the cumulative input exposure to the detector increases. Severe degradation in NPS was observed, which gives rise to reduction in DQE as the cumulative input exposure increases. With carefully updated offset and gain correction, however, we can overcome the detrimental effects of increased dark current on NPS and DQE. Index Terms CMOS detector, CMOS devices, detective quantum efficiency, digital radiography, flat-field correction, image evaluation, image sensors, noise-power spectrum, radiation effects, X-ray imaging. I. INTRODUCTION VOLUMETRIC computed tomography (CT) based on flatpanel detectors has recently become popular in diverse application fields such as image-guided radiotherapy [1], angiography [2] and small-animal imaging [3]. In every moment of any given imaging task and thus over the lifetime of a detector, deposition of a certain amount of radiation dose in an electronic pixel element is inevitable. This radiation effect can give rise to a change in the pixel-element characteristics such as the leakage current of a photodiode and the threshold voltage of a thin-film transistor (TFT) [4], [5], which may consequently degrade the performance and image quality of the imaging detector. While a large amount of effort has been invested in evaluating the performance degradation of the pixel-element devices themselves due to the radiation damage, there has not been sufficient attention paid to the evaluation of performance degrada- Manuscript received June 27, 2008; revised November 25, 2008. Current version published June 10, 2009. This work was supported by the Korea Research Foundation Grant ("KRF-2008-313-D01339) and the Basic Research Program of the Korea Science & Engineering Foundation (R01-2006-000-10233-0) funded by the Korean Government. *H. K. Kim was the corresponding author for this article. *H. K. Kim and M. K. Cho are with the Mechanical Engineering Department, Pusan National University, Busan 609-735, Korea (e-mail: hokyung@pusan.ac.kr; mk2000@pusan.ac.kr).. T. Achterkirchen is with Rad-icon Imaging Corp., Santa Clara, CA 95054 USA (e-mail: achterkirchen@rad-icon.com). W. Lee is with the Radiological Science Department, Korea University, Seoul, Korea (e-mail: wonhol@korea.ac.kr). Digital Object Identifier 10.1109/TNS.2009.2014232 tion of an imaging detector that is an important issue affecting the quality control and assurance of detectors over their lifetime. We have investigated the effects of radiation on the image quality of a CMOS (complementary metal-oxide-semiconductor) photodiode array detector coupled (via direct contact) to a phosphor screen. This CMOS detector is becoming popular in medical imaging applications because of its unique advantages, such as very low image lag and larger pixel fill factor compared to conventional flat-panel detectors based upon amorphous silicon TFT technology [6]. We have constructed an experimental setup to provide a large radiation dose to the CMOS detector, and measured dark signal and noise performance as a function of the radiation exposure. Based upon the measured dark signal and noise, we defined a parameter related to the dynamic range and investigated its degradation as a function of the cumulative input exposure. In addition, we measured the performance in terms of modulation-transfer function (MTF), noise-power spectrum (NPS) and detective quantum efficiency (DQE). We report the observation of radiation effects and discuss the role of offset and gain corrections on these metrics. II. MATERIALS AND METHODS A. Sample Detector Preparation We constructed a sample detector with a commercial phosphor screen (Min-R, Eastman Kodak, USA) and a CMOS photodiode array. The phosphor screen is mainly made up of terbium-doped gadolinium oxysulfide (Gd O S:Tb) and a polyurethane elastomer as the phosphor and binder, respectively. The thickness is 90 m and the density is 3.77 g/cm. The detailed physical configuration, parameters and specifications are described in [7]. The CMOS photodiode array has a format of 512 1024 pixels with a pitch of 48 m [8]. The photodiode array uses a source-follower configuration to read out the pixel voltage. The pixel consists of a photodiode to collect charge, a reset switch to apply a reverse-bias voltage to the photodiode, a row-select switch to select that pixel for read out, and a source-follower to convert the charge to voltage onto a column bus. This active pixel configuration, unlike the passive pixel design used in amorphous silicon TFT detectors, does not require an amplifier on each column because it is sufficient to sample the voltage on the column bus. In most cases, the gain of the pixel source-follower, i.e., the transfer gain from the photodiode voltage to the voltage on the column bus, is slightly less than unity. There are several source-follower stages in the signal processing chain. 0018-9499/$25.00 2009 IEEE

1180 IEEE TRANSACTIONS ON NUCLEAR SCIENCE, VOL. 56, NO. 3, JUNE 2009 The total conversion gain from the pixel to sensor output is 0.5 V/, and the pixel full-well capacity is approximately. The photodiodes in the CMOS photodiode array are always integrating, which means that it is always collecting charge. The readout mechanism is similar to the correlated double sampling (CDA) technique, which is widely used in CCDs (charge-coupled devices) to reduce switching transients and to eliminate reset noise. The signal values from an entire row of photodiodes are sampled onto a set of capacitors, then the photodiodes are reset, and then the reset values from the same pixels are sampled onto a second set of capacitors. The readout then proceeds by reading out the stored voltages from both sets of capacitors. Voltage signals from the CMOS photodiode array are then digitized to 12-bit resolution by an analog-to-digital converter (ADC). During the measurements, the frame time of the CMOS detector was fixed at 540 ms, although it is variable and the maximum frame rate is 4.5 frames per second (fps). B. Experimental and Image Data Analysis A sealed x-ray source (Series 5000 Apogee, Oxford Instruments, USA) with a fixed tungsten anode and a 125- m-thick beryllium exit window was used to irradiate the detector. In general, the radiation effect on a device is evaluated in terms of the total ionizing dose (or total dose), which is the accumulated energy per unit mass of the device [9]. It is, however, difficult to directly measure the actual energy absorbed in the device. Instead, we measured the exposure rate in Roentgens (R) at the detector entrance surface by using a calibrated ion chamber (Victoreen 6000-528, Inovision, USA). For the sample detector, the irradiation was performed using a 45-kVp x-ray spectrum tailored by a 1-mm-thick aluminum filter, which gives an exposure rate of 13 R/min measured at a distance of 200 mm from the source. The following measurements are reported as a function of cumulative input exposure at the entrance surface of the CMOS detector; in other words, the measured exposure rate multiplied by the elapsed irradiation time. Eight images were obtained to evaluate the dark signal and noise with respect to the cumulative input exposure. Each image was divided into eight non-overlapping sub-image sections, each consisting of 128 128 pixels. Thus a total of 64 region-of-interests (ROIs) were used for the evaluation. The increased dark signal in the detector can be corrected using the offset correction algorithm that is part of the typical digital x-ray imaging procedure. However, radiation-induced dark signal can cause a decrease in capacity of signal charge or bit-depth. The reduction of bit-depth is the loss of range available to image different anatomies with different attenuation values in medical imaging. Therefore, the reduction of signal capacity is crucial to the long-term use of a detector. The conventional definition of dynamic range is the ratio of the maximum signal divided by the noise of the signal. In order to investigate the reduction of signal capacity, instead we have defined a modified dynamic range as a function of the cumulative input exposure (1) where and are the maximum signal and the dark signal at cumulative input exposure, respectively. is the standard deviation of the dark signal at cumulative input exposure. The upper bar denotes the mean value. It should be noted that ADU in this study. The performance of the CMOS detector has been measured in terms of MTF, NPS, and DQE. A detailed description of measurement procedures can be found in the [6], [10]. Briefly, the presampling MTF was measured using a slanted-slit method to avoid aliasing [11] and the NPS was determined by two-dimensional (2D) Fourier analysis of white images [12]. The DQE was assessed from the measured MTF, NPS, and the estimated photon fluence. The fluence was calculated using the experimentally measured exposure and half-value layer (HVL), and a computational program for x-ray spectral analysis [13]. In this study, the NPS is calculated by where the vector form designates a spatial-frequency conjugate in Fourier domain to in space domain and indicates a 2D fast Fourier transformation of an image. is the mean value of the data within a subarray of a white image being investigated. On the contrary, a normalized NPS (NNPS), which is typically used for the evaluation of DQE because it does not require the additional measurement of the detector gain, is calculated by In order to investigate the effect of different gain-offset correction procedures on the detector performance, three different correction procedures were applied when obtaining an NPS or NNPS. In the first method, for an image investigated at a given cumulative input exposure, the image is corrected with initial dark and white images obtained just before the continuous x-ray irradiation such that The symbol denotes an ensemble average of images and denotes the process of taking a subarray image from the corrected white image. denotes the mean value of the offset corrected image in a restricted area, usually a central region of the detector. In diagnostic radiology with conventional flat-panel digital radiography (DR) detectors, the concept of gain-offset correction is similar to this correction procedure. Averaged dark and white images are first obtained when the DR system is initialized and are subsequently used for the gain-offset correction procedure during the daily diagnostic imaging tasks. In the second method, the correction is carried out with initial dark images obtained just before the continuous irradiation and (2) (3) (4)

KIM et al.: RADIATION EFFECTS ON THE PERFORMANCE OF CMOS PHOTODIODE ARRAY DETECTORS 1181 Fig. 1. Signal and noise properties in the CMOS detector during irradiation. (a) Dark signal of the CMOS detector as a function of the cumulative input exposure. (b) Log-log plot of noise versus dark signal of the CMOS detector. Noise property of difference images, which contain only random noise, is also plotted. refreshed white images obtained at the corresponding cumulative exposure level In the third method, the offset and gain correction is carried out with refreshed dark and white images: (5) III. RESULTS Fig. 1(a) shows the measured dark signal of the CMOS detector as a function of the cumulative input exposure. It is noted that the dark signal is expressed in ADC units (ADU). As the cumulative input exposure increases, the dark signal also increases. In Fig. 1(b) a log-log plot of noise versus dark signal demonstrates a signal-dependent noise contribution to signal values. As shown in Fig. 1(b), the signal-noise dependency can be categorized into three regions for the dark signal levels; a signal-independent or constant region, a square root proportional region and a linear proportional region. The constant region reflects the read noise of the detector, which initially dominates all other noise sources. As the radiation-induced dark current begins to increase, the noise also increases with a square-root dependence, since the dark current generation is a thermal process and thus follows Poisson statistics. Finally, the increased dark current becomes pronounced enough to form a fixed-pattern background, where the standard deviation increases linearly with the signal. It should be noted that this background can be subtracted out by periodically renewing the offset calibration, but for this illustration the same offset correction was applied to images taken at all cumulative input exposure levels. (6) Fig. 2. Dynamic range with respect to the cumulative input exposure. The dynamic range defined in this study decreases as the cumulative input exposure is increased, due to a gradual reduction in signal charge capacity as well as a simultaneous rise in noise levels. Dynamic range estimated based on difference images is also plotted. Noise of the difference of two dark images obtained at the same cumulative input exposure as a function of the dark signal level is also plotted in Fig. 1(b). Subtraction of two images obtained under identical condition produces an image free from fixed-pattern noise (FPN). According to the noise plot obtained from the difference images, as shown in Fig. 1(b), most of the noise in the dark images induced by irradiation is FPN. It is noted that the noise for each difference image was corrected by dividing by because the two images have independent noise samples with equal standard deviation. The calculated dynamic range defined in this study is plotted in Fig. 2 based on the measured dark signal and noise. As expected, the dynamic range decreases as the cumulative input exposure increases because of the increase of noise and the enhanced dark current. The dynamic range drops by a factor of 100 over the course of the experiment. However, based on the difference images, i.e., by removing the FPN, the drop is a factor of about 10. The reduction in dynamic range during the use of detectors is significant for any clinical imaging application for the obvious reason of loss of range available for the expected range of x-ray intensities after attenuation by human anatomy. Fig. 3 plots the measured MTF of the detector with respect to the cumulative input exposure. As shown in Fig. 3, the MTF

1182 IEEE TRANSACTIONS ON NUCLEAR SCIENCE, VOL. 56, NO. 3, JUNE 2009 Fig. 3. Measured MTFs of the CMOS detector with respect to the cumulative input exposure. MTF, which describes the resolving power of an imaging system, is rarely affected by the cumulative input exposure. The exceptional degradation at a cumulative input exposure of about 3.6 kr is caused by that the slit image was very noisy. Fig. 5. Normalized NPSs of the CMOS detector with respect to the cumulative input exposure for three different gain-offset correction procedures. Fig. 4. Measured NPSs of the CMOS detector with respect to the cumulative input exposure. Three different gain-offset correction procedures with initial dark and white images (a), initial dark and refreshed white images (b), and refreshed dark and white images (c) are applied. The NPS measures the change in the noise amplitude as a function of spatial frequency and bridges the noise and spatial resolution in an image. performance was very slightly affected by the cumulative input exposure except at the very highest level of about 3.6 kr. However, since the slit images obtained for the MTF evaluation became more noisy as the cumulative input exposure increased, it was difficult to analyze whether the degradation is due to the irradiation or not. The measured NPS curves with respect to three different correction methods are plotted in Fig. 4. For sequential images obtained during continuous irradiation, gain-offset correction with the initial dark and white images obtained just before irradiation (or kr of the relative cumulative exposure) produces incremental noise-power densities in NPS as the cumulative exposure increases. This is due to the addition of increased dark current dependent on the amount of irradiation in white images. Since the fixed-pattern background in the dark and white images is getting dominant as the cumulative exposure increases, the improper gain-offset correction may yield a noticeable pattern in the NPS. As shown in Fig. 4(a), peaks at the low spatial-frequency region are observed. The FPN in this study is then characterized as long-range non-uniformities in an image. On the other hand, regardless of the use of initial or refreshed dark images, gain-offset corrections with refreshed white images obtained at the corresponding exposure produce the consistent NPS results as shown in Fig. 4(b) and (c). The importance of the use of proper dark images in the gainoffset correction procedure can be observed by investigating the normalized NPS curves. As shown in Fig. 5(b), normalized spectral densities in NNPSs based on the refreshed white but initial dark images decreases as the cumulative input exposure increases because increases as the cumulative input exposure increases. Therefore, only the gain-offset correction procedure with the refreshed dark and white images can produce the consistent NPS and NNPS results. From the measured MTF and NNPS, the resultant DQE of the detector with respect to the cumulative input exposure is shown in Fig. 6. In order to match the spatial frequencies between the MTF and NNPS, we fitted the MTF data with a Lorenzian function. The DQE at the highest level of dose of about 3.6 kr was excluded in the calculation because of the inconsistent MTF data. As shown in Fig. 6, DQE is mainly dependent on the NNPS. IV. DISCUSSION The measured dark signal depends on the photodiode, the transistors and the interconnectivity in a pixel [14]. Among them, the contribution of the photodiode dark current is highest. The main physical mechanism responsible for the increasing dark current in the photodiode is the buildup of positive charge in the silicon dioxide (SiO ) layer. When an x-ray photon is

KIM et al.: RADIATION EFFECTS ON THE PERFORMANCE OF CMOS PHOTODIODE ARRAY DETECTORS 1183 providing the electrons in the silicon substrates with enough energy to jump back into the SiO layer, for example by using thermal energy as the activation energy. As shown in Fig. 7, while room-temperature annealing did not work to restore the dark signal, we have had some success with annealing the damaged CMOS photodiode array by baking it in a vacuum oven at a temperature around 100 C for 58 hours, while taking precautions not to degrade the materials inside the array at higher temperatures. If we can apply higher temperatures, the deeper traps could be relieved. Fig. 6. DQEs of the CMOS detector with respect to the cumulative input exposure for three different gain-offset correction procedures when estimating NPSs. The DQE measures the fraction of incident fluence contributing to an image quality. V. CONCLUSION We have measured the radiation damage and its effect on the performance of a CMOS photodiode array in conjunction with a phosphor screen in terms of dark signal, noise, dynamic range, MTF, NPS and DQE. For the accumulation of the radiation exposure, the radiation-induced increase both in dark signal and noise resulted in the gradual reduction in dynamic range. Most noise in dark images induced by irradiation is a fixed-pattern background. Although the radiation-induced noise can affect the NPS or NNPS, and hence DQE, the proper gain-offset correction with carefully updated dark and white images can prevent this type of degradation. This study gives an initial motivation that the periodic monitoring of the performance degradation is an important issue for the long-term and healthy use of digital x-ray imaging detectors. Fig. 7. Effect of annealing on the restoration of signal capacity in the CMOS detector. The hatched area indicates an occasional forced heat-treatment in a vacuum oven at around 100 C for 58 hours. absorbed in the SiO layer, it can accelerate electrons out of the SiO and into the silicon substrate. Since the SiO is a dielectric, it cannot replace the missing electrons. Consequently, over time, a positive charge builds up in the SiO. This charge distribution gives rise to an electric field that modifies the band levels in the active devices. In other words, positive charges in the SiO cause electrons to accumulate underneath the surface, modifying the charge density in the depletion region and thereby increasing the leakage current across the - junction of the photodiode. The dark current becomes stronger with increasing absorbed dose. For radiation-induced dark signal, Bogaerts and Dierickx [14] simply modeled that the thermal generation predominantly follows from the creation of interface traps. From the measurements in this study, we found that the dark signal almost linearly increased up to a cumulative input exposure of kr, and that increase of the dark signal accelerated even further beyond kr. Radiation-induced interface traps in the SiO layer are in general positive ions. Some missing electrons can be restored by REFERENCES [1] D. A. Jaffray and J. H. Siewerdsen, Cone-beam computed tomography with a flat-panel imager: Initial performance characterization, Med. Phys., vol. 27, no. 6, pp. 1311 1322, 2000. [2] R. Ning, B. Chen, R. Yu, D. Conover, X. Tang, and Y. Ning, Flat panel detector-based cone-beam volume CT angiography imaging: System evaluation, IEEE Trans. Med. Imag., vol. 19, no. 9, pp. 949 963, 2000. [3] S. C. Lee, H. K. Kim, I. K. Chun, M. H. Cho, S. Y. Lee, and M. H. Cho, A flat-panel detector based micro-ct system: Performance evaluation for small-animal imaging, Phys. Med. Biol., vol. 48, no. 24, pp. 4173 4185, 2003. [4] J. M. Boudry and L. E. Antonuk, Radiation damage of amorphous silicon photodiode sensors, IEEE Trans. Nucl. Sci., vol. 41, no. 4, pp. 703 707, 1994. [5] J. M. Boudry and L. E. Antonuk, Radiation damage of amorphous silicon, thin-film, field-effect transistors, Med. Phys., vol. 23, no. 5, pp. 743 754, 1996. [6] H. K. Kim, S. C. Lee, M. H. Cho, S. Y. Lee, and G. Cho, Use of a flat-panel detector for microtomography: A feasibility study for smallanimal imaging, IEEE Trans. Nucl. Sci., vol. 52, no. 1, pp. 193 198, 2005. [7] J.-M. Kim, H. K. Kim, M. H. Cheong, M. K. Cho, C.-S. Shon, and C. H. Lim, Investigation of scintillation screens for x-ray imaging, Key Eng. Mater., vol. 321 323, pp. 1056 1059, 2006. [8] T. Graeve and G. Weckler, High-resolution CMOS imaging detector, in Proc. SPIE, 2001, vol. 4320, pp. 68 76. [9] H. J. Barnaby, Total-ionizing-dose effects in modern CMOS technologies, IEEE Trans. Nucl. Sci., vol. 53, no. 6, pp. 3103 3121, 2006. [10] M. K. Cho, H. K. Kim, T. Graeve, and J.-M. Kim, Characterization of CMOS pixel detectors for digital x-ray imaging, Key Eng. Mater., vol. 321 323, pp. 1052 1055, 2006. [11] H. Fujita, D. Y. Tsai, T. Itoh, K. Doi, J. Morishita, K. Ueda, and A. Ohtsuka, A simple method for determining the modulation transfer function in digital radiography, IEEE Trans. Med. Imag., vol. 11, no. 1, pp. 34 39, 1992. [12] E. Samei and M. J. Flynn, An experimental comparison of detector performance for direct and indirect digital radiography systems, Med. Phys., vol. 30, no. 4, pp. 608 622, 2003. [13] C.-S. Shon, H. K. Kim, M. K. Cho, M. H. Cheong, C. H. Lim, and J.-M. Kim, Computational toolset for x-ray spectral analysis, Key Eng. Mater., vol. 321 323, pp. 1060 1065, 2006. [14] J. Bogaerts and B. Dierickx, Total dose effects on CMOS active pixel sensors, in Proc. SPIE, 2000, vol. 3965, pp. 157 167.